基于混合代理建模的人工人口生成与标定方法研究

Multi-agent based artificial population has become fundamental to the analysis of various complex systems such as urban transportation, regional economy, military war, infectious disease and so on. Traditional micro population simulation concentrates on the basic data fitting, whereas agent based social simulation pays more attention to model human decision-making process. This project tends to integrate advantages of the two approaches, and give a research about the generation and calibration methods of artificial population based on hybrid agent modeling. Improved basic population database synthesis method will be firstly presented, and agent decision model as well as behavioral rules will be designed then by integrating the sociality, norms, and emotional personality of humans according to the DISC (Dominance, Influence, Steadiness, Compliance) personality theory. In the following, the project will explore the calibration method of agent model parameters using Markov Chain which bridges a link between micro agent decision parameters and macro population distribution. At last, we will adopt national census and annual demographic survey data as our input to construct an artificial population simulation system, study the evolution of the basic demographic attributes, and validate our population generation and calibration methods. This research will consolidate the realistic foundation of artificial population, leading the analysis of concerned complex systems more reliable and practicable.

基于多代理的计算机人工人口是分析城市交通、区域经济、军事战争、疾病传播等各类复杂系统的重要基础。由于传统的微观人口仿真强调基础数据拟合，而基于代理的社会仿真则注重于对人的决策过程建模，本项目拟综合二者优势，开展基于混合代理的人工人口生成与标定研究。项目拟首先给出不依赖于样本的完善的基础人口合成方法，并参考DISC(Dominance, Influence, Steadiness, Compliance)性格理论，综合考虑代理的社会性、性格情感和行为规范设计代理的决策模型和行为规则，然后采用马尔科夫链将代理的微观决策参数和宏观分布特征相结合，给出代理模型参数的标定方法，最后以我国人口普查和年度调查数据为输入构建人工人口计算模拟系统，研究基本人口属性的演化趋势并验证人口合成与参数标定方法的正确性。该项研究有助于进一步提升人工人口系统的现实基础，使相关复杂系统的研究与分析更加具有可靠性和实用性。

基于多代理的计算机人工人口是分析城市交通、区域经济、军事战争、疾病传播等各类复杂系统的重要基础。由于传统的微观人口仿真强调基础数据拟合，而基于代理的社会仿真则注重于对人的决策过程建模，本项目综合二者优势，开展了基于混合代理建模的人工人口生成与标定研究。项目构建了基础人口合成、认知决策建模、代理模型标定三个步骤的人工人口建模范式，并分别提出了基于总体分布估计的大规模人口合成方法、代理通用认知决策模型和基于马尔科夫链状态转移的代理模型标定方法。与现有主流技术相比，所提方法能在精度不变的前提下更快速地合成大规模人口，并且人口动态模拟跟踪实际系统的误差更小。相关技术在人口迁移、城市交通管理等领域得到示范应用，并在青岛、杭州等地的城市智能交通改造中取得了良好的效果。